Fix usage & billing for custom model aliases and cached/reasoning tokens

[melihmucuk]

Summary:

• Ensure custom model aliases inherit the real model metadata so context/limits and cost are computed from the actual model.
• Correct usage/cost calculation: bill cached/uncached input tokens separately, include cache-write tokens, and bill reasoning tokens at the same rate as output tokens.

Changed files:

• packages/opencode/src/provider/provider.ts — fallback to lookup parsed.models[model.id] for aliases.
• packages/opencode/src/session/index.ts — getUsage: compute uncachedInput = inputTokens - cachedInputTokens, bill cached/uncached input separately, include cache.write, and charge reasoning tokens at output rate.

Tests: ran session tests locally (2 passed).

Fixes #4162

[melihmucuk]

@rekram1-node, we fixed the same bug, but this also fixes the spent calculations.

[rekram1-node]

@melihmucuk can u explain the cached input stuff? Cached input isn't free

[melihmucuk]

@rekram1-node Yes, cached input isn’t free. The SDK returns total input tokens, but we need to know how many of those were uncached to calculate spent correctly.

For example, suppose the SDK reports 1M input tokens: 100K uncached and 900K cached. We can’t calculate spent as 1M * input_token_price. We have to separate cached and uncached input tokens and then calculate the cost accordingly.

[rekram1-node]

okay cool, I think I misread ur original comment which got me confused so apologies

[rekram1-node]

@melihmucuk can u resolve the conflict, I think ur changes are correct for the input tokens

[melihmucuk]

@rekram1-node done.

[melihmucuk]

@rekram1-node ok for revert, but definitely add reasoning token cost to total cost. Almost all providers calculate the reasoning tokens as output tokens.

[rekram1-node]

Yeah for sure ill add in

[rekram1-node]

ill merge this in sorry got distracted